MOBILE POINT CLOUD SEMANTIC SEGMENTATION USING ARTIFICIAL NEURAL NETWORK AND APPROPRIATE PARAMETER DETERMINATION

The processing and information extraction of mobile point clouds has become an essential field of study in photogrammetry, remote sensing, computer vision, and robotics. Semantic segmentation is called to evaluate the singular features of the points together and collect them under meaningful clusters. This study aims to perform semantic segmentation with appropriate parameter selection using artificial neural networks. In addition, a study has been carried out to optimally define a point in the point cloud with the different feature spaces produced. Accordingly, eigen-based features are defined for each point. Eigen-based features describe the local geometry around the point and are commonly used in LiDAR processing today. Then, the most suitable parameters for semantic segmentation are determined. Multilayer Perceptron (MLP), an artificial neural network approach, was used in the study. The multilayer perceptron (MLP) is an artificial neural network to train any given non-linear input and contains several layers. Therefore, MLP is a suitable approach for solving non-linear problems. MLP has three layers: the input layer, the hidden layer, and the output layer. Paris-Carla-3D MLS dataset was used in the study. Paris-Carla-3D consists of two datasets, real (Paris) and synthetic (Carla). The dataset consists of data collected on a route 550 meters in Paris, 5.8 km in CARLA. The only real part Paris was used in this study. The highest mIoU metrics were obtained as 21.85% with the 0.4 m support radius, 30000 training samples and 200 hidden layer size.